🤖 AI Summary
To address the challenge of jointly modeling node embeddings and graph structure evolution in temporal interaction dynamic graphs, this paper proposes Graph Neural Controlled Differential Equations (GN-CDE), the first application of neural controlled differential equations to dynamic graph representation learning. GN-CDE employs continuous-time dynamics to jointly characterize both node embedding trajectories and graph topology evolution, enabling end-to-end, piecewise-integration-free continuous evolution modeling, posterior trajectory calibration, and robust estimation under missing observations. The method integrates continuous-time graph neural networks, explicit time-derivative modeling, and differentiable ODE solvers. Extensive experiments on multiple dynamic graph prediction and classification tasks demonstrate that GN-CDE significantly outperforms state-of-the-art baselines, validating the effectiveness, generalizability, and robustness of continuous dynamical modeling for temporal graph representation learning.
📝 Abstract
This paper focuses on representation learning for dynamic graphs with temporal interactions. A fundamental issue is that both the graph structure and the nodes own their own dynamics, and their blending induces intractable complexity in the temporal evolution over graphs. Drawing inspiration from the recent process of physical dynamic models in deep neural networks, we propose Graph Neural Controlled Differential Equation (GN-CDE) model, a generic differential model for dynamic graphs that characterise the continuously dynamic evolution of node embedding trajectories with a neural network parameterised vector field and the derivatives of interactions w.r.t. time. Our framework exhibits several desirable characteristics, including the ability to express dynamics on evolving graphs without integration by segments, the capability to calibrate trajectories with subsequent data, and robustness to missing observations. Empirical evaluation on a range of dynamic graph representation learning tasks demonstrates the superiority of our proposed approach compared to the baselines.
